Upload
duongtuong
View
227
Download
0
Embed Size (px)
Citation preview
Hadley Centre Technical Note 96Disaggregation of daily data in JULES
July 23, 2014
Karina Williams, Met Office, and Douglas Clark, Centre for Ecologyand Hydrology
1 Introduction
The Joint UK Land Environment System (JULES) [Best et al., 2011, Clark et al., 2011] is a com-
munity land surface model that can be used both online as part of the Met Office Unified Modelling
system and offline for impacts studies. It has been designed to run at short time scales, in order to
exchange heat, momentum and moisture fluxes with an atmospheric model at model timesteps.
When run in offline mode, the meteorological data used to drive JULES is read from input files.
This forcing data is often only available at a temporal resolution of three hours or coarser, due
to the practical considerations in storing such a large amount of atmospheric model data and the
difficulties of obtaining observational data at this resolution. JULES has been shown to be sensitive
to the temporal resolution of the forcing data (e.g. Compton and Best [2011]).
In some cases, particularly for projects involving model intercomparisons (e.g. the Inter-Sectoral
Impact Model Inter- comparison Project (ISI-MIP) [Warszawski et al., 2013] and EUPORIAS [Hewitt
et al., 2013]), forcing data is only available at daily resolution. The absence of a diurnal cycle
for temperature and radiation and any sort of sub-daily structure for precipitation makes this data
completely unsuitable for forcing JULES directly. It is therefore necessary to perform some sort
of data disaggregation. To date, two main approaches have been used to disaggregate daily data
into sub-daily data for JULES: the WATCH [Weedon et al., 2011] approach, developed as part of
the EU-funded Water and Global Change programme, and the IMOGEN approach, used in the
IMOGEN modelling system [Huntingford et al., 2010] and the JULES runs for the ISI-MIP project
[Davie et al., 2013, Dankers et al., 2013]. The WATCH project published standalone fortran routines
to perform the disaggregation, whereas, in IMOGEN, the disaggregation is performed internally,
as the IMOGEN configuration of JULES is running. These approaches differ significantly in their
treatment of precipitation, with the WATCH disaggregator using a multiplicative cascade model to
distribute the precipitation and the IMOGEN approach modelling the days precipitation as arising
from one event of constant intensity and globally specified duration.
In JULES 4.0, the IMOGEN approach to disaggregation is made available in the JULES trunk
for standard (i.e. non-IMOGEN) runs. This report documents the method used and illustrates the
dependence on the value of the convective rain duration parameter chosen by the user, using a
development branch based on adding the disaggregation code to a branch of JULES 3.4.1, which
has been merged with the JULES trunk.
2 Disaggregation method
In JULES 4.0, disaggregation of daily forcing data to JULES model timesteps (typically of the order
of 30 minutes) can be switched on using the l_daily_disagg flag (see the JULES 4.0 manual for
more details of the user interface). When this flag is set to True, the daily forcing data is disaggre-
gated to model timesteps, which involves imposing a diurnal cycle on temperature and radiation and
c Crown Copyright 2014 1
allocating the precipitation to a continuous series of model timesteps within the day. The pressure
P and wind values are unchanged by the disaggregation code and the specific humidity is kept
below the specific humidity at saturation. Any interpolation (specified by interp in JULES DRIVE)
is performed within JULES before the disaggregation code is called. The disaggregation of each
variable is described in more detail in the following sections. The forcing variables required when
l_daily_disagg=T are the same as those needed when l_daily_disagg=F, plus the temperature
range for each grid box for each day.
2.1 Temperature
A diurnal cycle is imposed on the near-surface air temperature using the relation
T = T0 +T
2
cos(2 (t tT
max
) /t
day
), (1)
where T0 and T are the temperature and diurnal temperature ranges before disaggregation
respectively. tday
is the length of a day. tT
max
is the time of day where the temperature is highest
and is calculated with the IMOGEN routine sunny, which assumes that tT
max
occurs 0.15 of a
daylength after local noon i.e.
t
T
max
=
t
up
+ t
down
2
+ 0.15 (t
up
tdown
) (2)
where tup
and tdown
are sunrise and sunset times.
2.2 Specific humidity
The disaggregator code also ensures that the specific humidity q does not go above the specific
humidity at saturation qsat
(T, P ), as calculated by the subroutine qsat i.e.
q = min(q, q
sat
(T, P )). (3)
Note that this condition does not conserve water 1.
2.3 Radiation
The diurnal cycle imposed on the downward longwave radiation is a function of temperature, as
in Huntingford et al. [2010], derived assuming black body radiation and that the diurnal cycle of
temperature is a small perturbation,
R
down
lw
= R
down
lw,0
4
T
T0 3
, (4)
1After the model runs for this report were carried out, an extra option was added which keeps the relative humidityconstant - see the manual for more information.
c Crown Copyright 2014 2
where Rdownlw,0 is the downward longwave radiation before disaggregation and Rdownlw is the down-
ward longwave radiation after disaggregation.
The downward shortwave radiation is found by
R
down
sw
= R
0,downsw
R
norm
(5)
where Rnorm
is the solar radiation normalisation factor, calculated by the IMOGEN routine sunny,
using the position of the sun in the sky at each timestep for each gridbox. Rdownsw,0 and Rdownsw are the
downward shortwave radiation before and after disaggregation, respectively.
In a general run of JULES, the diffuse radiation can be given explicitly in the forcing data (this is
not the case for IMOGEN runs). A diurnal cycle is imposed on this diffuse radiation using
R
diff
= R
0diff
R
norm
(6)
2.4 Precipitation
The IMOGEN method of disaggregating the precipitation has two components:
The entire days precipitation of each type is first restricted to one event, of duration x
for
x = convective rain, large-scale rain, convective snow and large-scale snow respectively (in
seconds). The start of the precipitation event is distributed randomly throughout the time
period between the beginning of the day (in UTC) and x
before the end of the day.
If the rate of precipitation in a timestep exceeds a maximum precipitation rate (hardwired in as
350 mm/day), then the precipitation is redistributed using the IMOGEN routine redis, which
moves the excess precipitation to dry timesteps immediately before or after the sequence of
wet timesteps, thus lengthening the event beyond x
. This step was included in IMOGEN
because it was found that very high precipitation rates caused numerical issues for the Met
Office Surface Exchange Scheme (MOSES) Essery et al. [2001], which evolved into JULES
(for more details, see code comment in src/imogenday calc.F90).
The mean diurnal cycle produced by this method is not a constant. Since rain events are not
allowed to begin in one day and finish in the subsequent day, the timesteps near the beginning and
end of the day (as defined by the time used internally in JULES, not by local time) will be wet less
often than timesteps in rest of the day. This is illustrated in Figure 1, which shows the mean diurnal
cycle for 100,000 days of precipitation for a range of event durations.
There are four options for precipitation disaggregation in JULES 4.0, specified by the value of
the precip disagg method flag:
1. a constant value of precipitation rate is used for the entire day
2. the IMOGEN method
c Crown Copyright 2014 3
Figure 1: The diurnal cycle of simulated precipitation rate for 100,000 days, each with a mean dailyprecipitation rate normalised to one. The precipitation for each day has been disaggregated into 72timesteps i.e the timestep length is 20 minutes. The legend shows the precipitation event durationin hours.
3. as for the IMOGEN method except precipitation does not get redistributed when the rate ex-
ceeds max precip rate
4. a proportion x
/t
day
of the timesteps in the day are randomly selected to be wet, and the
precipitation is distributed uniformly among them (tday
is the length of a day in seconds). This
enforces a flat diurnal cycle of precipitation.
Global values of the precipitation event durations (x
) are set by the user (in JULES DRIVE). The
default values are conv rain
= 6 3600 seconds, ls rain
=
ls snow
= 1 3600 seconds. These have
been taken from the defaults in IMOGEN at JULES 3.4.1. For ISI-MIP, conv rain
= 2 3600 seconds,
ls rain
=
ls snow
= 5 3600 seconds, chosen to be physically reasonable. Convective snow was
not given as input in IMOGEN or ISI-MIP. The duration of convective snow events has a default in
JULES 4.0 of conv rain
= 1 3600 seconds.
3 Experimental setup
The performance of this disaggregation method was investigated using output from a global vn8.3
Met Office Unified Model (UM) run in the Global Atmosphere 4.0 (GA4.0) configuration [Walters
et al., 2013] at a model timestep of 20 minutes as forcing data for an offline run of a branch of
JULES 3.4.1 which had the disaggregation code added. This branch was merged with the JULES
trunk before the release of JULES 4.0. The longwave radiation was pre-processed by copying the
data value at the first timestep of each hour into the second and third timesteps of that hour, in order
to eliminate errors that had been introduced by requiring model output at a higher resolution than
the radiation timestep. The JULES namelists for the offline runs were based on the GL4.0 JULES
configuration namelists, edited to be consistent with the UM run, including use of the same ancillary
c Crown Copyright 2014 4
files.
The results from comparing an offline JULES run driven by this 20 minute forcing data was
compared to the results when JULES was driven by daily means of this forcing data, which are
disaggregated internally using precip disagg method = 3 (i.e. similar to IMOGEN method but with
no maximum precipitation rate being enforced) and the diurnal temperature range and a range of
convective rain event durations conv rain
. For comparison, an additional run was done with three
hourly driving data with interpolation (interp = f).
4 Results
4.1 Diurnal cycle
Fig 2 illustrates the mean diurnal cycle imposed by the disaggregator on the driving data, averaged
over a year of the run (1984), as compared to the diurnal cycle of the model, using the Manaus
FLUXNET site in Brazil as an example (see Appendix for a selection of other FLUXNET sites). As
shown by Eq. 1 and Eq. 2, the disaggregator imposes a sinusoidal diurnal cycle on the temperature,
which in general works reasonably well for the middle of the day but has a minimum earlier than the
original 20 minute driving data; in the example of the Manaus site, this mimimum occurs four hours
too early. The mean short wave diurnal profile is captured well by the disaggregator. However, the
long wave diurnal profile of the original 20 minute forcing data has a much smaller amplitude than
produced by the disaggregator using Eq. 4, and, as was the case for the temperature, the amount
of time between the minimum and the following maximum in the original 20 minute driving data is
less than would be expected if the diurnal cycle followed a sine curve. As stated in Section 1, no
diurnal cycle is imposed on wind or pressure by the disaggregator. The specific humidity diurnal
cycle is also not well matched by the disaggregator.
The convective rainfall produced by the disaggregator at the Manaus site is a good example of
the limitations of the precipitation disaggregation method. Fig 2 shows disaggregated convective
rainfall with a default event duration of 6 hours. In the original 20 minute driving data, there is a clear
diurnal cycle. This is not reproduced at all in the disaggregated convective rainfall, which instead
tends to roughly a trapezoidal pulse, with minimum at midnight UTC, as discussed in Section 2.
Similarly, large-scale rainfall in the original 20 minute driving data has a diurnal cycle which is not
well described by the disaggregator, using the default large-scale rain event duration value of one
hour. If data from more years were included in this plot, the diurnal profile of the disaggregated
large-scale rainfall between 01:00 UTC and 23:00 UTC would tend towards a flat line.
In addition to plotting the mean diurnal cycle, it would also be useful to investigate how well the
variability is captured by the disaggregator. For example, short wave radiation would have variability
introduced by sporadic cloud cover, which is not modelled by the disaggregator.
c Crown Copyright 2014 5
Figure 2: Diurnal cycle of driving data for the Manaus FLUXNET site, Brazil, averaged over the daysin 1984.
c Crown Copyright 2014 6
4.2 Global climatology
Figure 3 (top left) shows the total evapotranspiration (total ET) over the full JULES offline run, which
has forcing data at 20 minute resolution created by the UM run (described in Section 3). Figure 3
(bottom right) shows the anomaly from subtracting this result from a run forced with daily means
and no disaggregation. Unsurprisingly, given the complete lack of diurnal cycle in this forcing data,
these anomalies are very large, particularly in the tropics.
Figure 3 (bottom left) shows the anomaly in total evapotranspiration from a run forced with daily
means and disaggregated internally in JULES to the model timestep with respect to the run driven
by the full 20 minute UM output. The event durations have been kept at their default values. It can
be seen that the use of the disaggregator reduces the anomalies drastically, implying that there is
a significant advantage to using the disaggregator if only daily forcing data is available. However,
this should be considered in the context of the anomalies from using 3 hourly forcing data with
interpolation Figure 3 (top right), which are larger than the anomalies from using the disaggregator.
Since 3 hourly forcing data with interpolation represents the model diurnal cycle and variability far
better than this disaggregator can, this implies that the small anomalies in Figure 3 (bottom right)
are perhaps due to a tuning of the disaggregator parameters to partially compensate for model
biases.
There is also some compensation between the contribution of soil ET anomalies and canopy ET
anomalies towards total ET, for example in South East Asia (Figure 7 and Figure 8 in the Appendix).
The anomalies in the soil evapotranspiration in the equatorial regions are predominantly negative in
the run driven by 3 hour means and daily means with no disaggregation, particularly in the tropics,
and the anomalies in the soil evapotranspiration are predominantly positive.
Figure 4 shows the mean runoff in the full run (top left) and the anomalies with the respect to
the full run for the 3 hourly-forced run (top right), the disaggregated run (bottom left) and the daily
forced run without disaggregation (bottom right). Comparison with Figure 3 (bottom left) shows that
the regions with large negative anomalies roughly coincide with the regions where the total evapo-
transpiration anomaly was high, as we would expect since the extra water added to the atmosphere
through evaporation and transpiration is not available for inclusion in the runoff. Figure 4 also shows
that the three hourly forced run and daily run have total runoff anomalies with a higher magnitude
than the disaggregated run, also consistent with the total evapotranspiration results. In some areas,
the contributions to the total runoff anomaly from surface runoff and sub-surface runoff act to com-
pensate for eachother to a certain extent (Figure 9 and Figure 10 in the Appendix), such as in the
region south and east of the Himalayas, where the disaggregated run shows a positive anomaly in
the surface runoff and a negative anomaly in the sub-surface runoff.
c Crown Copyright 2014 7
18
0
120
60
060
120
180
60
3003060
tota
leva
potr
ansp
irat
ion
0.0
0.8
1.6
2.4
3.2
4.0
4.8
mm
/day
18
0
120
60
060
120
180
60
3003060
tota
leva
potr
ansp
irat
ion
anom
aly,
3hrl
y,in
terp
olat
ed
1.
5
1.0
0.
50.
00.
51.
01.
5
mm
/day
18
0
120
60
060
120
180
60
3003060
tota
leva
potr
ansp
irat
ion
anom
aly,
disa
gg,
conv
rain
=6h
1.
5
1.0
0.
50.
00.
51.
01.
5
mm
/day
18
0
120
60
060
120
180
60
3003060
tota
leva
potr
ansp
irat
ion
anom
aly,
daily
,no
disa
gg
1.
5
1.0
0.
50.
00.
51.
01.
5
mm
/day
Figu
re3:
Top
left:
Out
putv
aria
ble
from
JULE
Sru
nw
ithfu
ll20
min
ute
reso
lutio
nfo
rcin
g.To
prig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
g3
hour
lyfo
rcin
gw
ithin
terp
olat
ion.
Bot
tom
left:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gdi
sagg
rega
tion
with
defa
ulte
vent
dura
tions
.B
otto
mrig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gda
ilyfo
rcin
gw
ithno
inte
rpol
atio
n.
c Crown Copyright 2014 8
18
0
120
60
060
120
180
60
3003060
tota
lrun
off
0.0
1.6
3.2
4.8
6.4
8.0
9.6
mm
/day
18
0
120
60
060
120
180
60
3003060
tota
lrun
offa
nom
aly,
3hrl
y,in
terp
olat
ed
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
18
0
120
60
060
120
180
60
3003060
tota
lrun
offa
nom
aly,
disa
gg,
conv
rain
=6h
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
18
0
120
60
060
120
180
60
3003060
tota
lrun
offa
nom
aly,
daily
,no
disa
gg
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
Figu
re4:
Top
left:
Out
putv
aria
ble
from
JULE
Sru
nw
ithfu
ll20
min
ute
reso
lutio
nfo
rcin
g.To
prig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
g3
hour
lyfo
rcin
gw
ithin
terp
olat
ion.
Bot
tom
left:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gdi
sagg
rega
tion
with
defa
ulte
vent
dura
tions
.B
otto
mrig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gda
ilyfo
rcin
gw
ithno
inte
rpol
atio
n.
c Crown Copyright 2014 9
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=3h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=6h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=12h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=24h
1.5 1.0 0.5 0.0 0.5 1.0 1.5mm/day
total evapotranspiration
Figure 5: Anomalies with respect to the full result, using disaggregation with convective rain eventdurations of 3 hours (top left), 6 hours (top right), 12 hours (bottom left) and 24 hours (bottom right).
4.3 Dependence on the convective rain event duration
Figure 5 shows the anomalies of the disaggregated runs with respect to the full run for convective
rain event duration parameters of conv rain
= 3h, 6h, 12h and 24h. Event duration parameters
for the other types of precipitation are kept at their default values. The total evapotranspiration
increases as conv rain
increases, as we would expect, as larger values of conv rain
means that the
rainfall is less intense and lasts longer.
As the convective rain event duration increases, it is generally the case that soil ET decreases
and canopy ET increases (Figure 11 and Figure 12 in the Appendix), as expected since the lower
event durations give more intense rain, which penetrates the canopy better. In some regions, such
as the Pampas in South America, which is predominantly modelled as C3 grass, soil evaporation
increases as the convective rain event duration increases. For total ET, soil ET and canopy ET, using
the disaggregator with the disaggregation of convective rainfall switched off (i.e. conv rain
= 24h)
is a significant improvement on using no disaggregation at all, illustrating the importance of the
disaggregation of the other driving variables. Out of this selection of values for the convective
rain event duration, conv rain
= 6h results in the lowest anomalies for total evapotranspiration, soil
c Crown Copyright 2014 10
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=3h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=6h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=12h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=24h
0.9 0.6 0.3 0.0 0.3 0.6 0.9mm/day
total runoff
Figure 6: Anomalies with respect to the full result, using disaggregation with convective rain eventdurations of 3 hours (top left), 6 hours (top right), 12 hours (bottom left) and 24 hours (bottom right).
evapotranspiration and canopy evapotranspiration in general.
The runoff also depends strongly on the duration of the convective rain events, shown in Figure
6. As for evaporation, conv rain
= 6h appears to perform the best for total runoff out of the values
of conv rain
illustrated here. As the duration decreases (and so the rainfall intensity increases),
the surface runoff increases and sub-surface runoff decreases (Figure 13 and Figure 14 in the
Appendix), as we would expect. In some regions, such as the Andes and the region south and east
of the Himalayas for conv rain
= 3h, the negative anomaly in sub-surface runoff roughly mirrors the
positive surface runoff anomaly. In other regions, such as the Congo basin for conv rain
= 12h, there
is a large negative anomaly in surface runoff, but no correspondingly large anomaly in sub-surface
runoff, leading to a large negative anomaly in the total runoff, consistent with the large positive
anomaly in total evapotranspiration, which we saw in Figure 5 (top right).
In summary, in it generally the case that increasing the convective rain event duration parameter
results in more of the rainfall being evaporated and less contributing to the runoff, particularly in
equatorial regions. Evaporation and runoff are very sensitive to this parameter, which should be set
with care. From the results shown here, using conv rain
= 6h appears to achieve a good compro-
c Crown Copyright 2014 11
mise for most areas of the world. However, selecting the values of the event durations by tuning the
evaporation and runoff anomalies could hide biases in the climatology of the model. There is good
indication that this is happening here, since, as we saw in Section 4.2, using conv rain
= 6h can
result in lower anomalies than using 3 hourly forcing data with interpolation.
5 Summary
A disaggregation option has been built in to JULES 4.0, which will allow offline runs to be forced
by daily meteorological data, which is then disaggregated to the model timestep internally using the
IMOGEN method. This option will be particularly useful for model intercomparisons, where subdaily
forcing data is not always available. We describe the implementation of this method in detail and
show an example of the resulting mean diurnal cycle of forcing variables after disaggregation.
We used UM output to force a set of offline JULES runs, to compare forcing at 20 minute res-
olution with forcing at three hourly resolution with interpolation and daily forcing with and without
disaggregation. There was a clear advantage to using the disaggregator when using daily driving
data. However, the disaggregated run with default parameters also reproduced the mean climate in
some cases better than the 3-hourly forced run, which implies that these parameter values may be
partially compensating for biases in the model. Since the results are sensitive to the disaggregation
parameters, we would encourage the user to consider carefully which values are most appropriate
for their individual needs and possibly to perform some initial analysis to assess the impact of this
choice on the region, output variables and timescales they are most interested in.
An interesting next step would be to investigate the variability of the model output produced using
the disaggregation option. It would also be useful to see how the disaggregator performs compared
to high temporal resolution observational forcing data since, for example, there are known issues
with the model diurnal cycle of precipitation [Stephens et al., 2010, Stratton and Stirling, 2012,
Stirling and Stratton, 2012].
There are many ways in which this disaggregator could be improved, which we hope will be
taken forward by the JULES community. For example, one simple extension would be to allow
the precipitation event duration parameters to vary spatially, although the simplistic nature of the
precipitation model (all the rainfall in one event, with constant intensity) means that it is not clear how
the duration parameter could be derived from observed precipitation, even when these observations
are available at sufficiently high temporal resolution. Other interesting possibilities would be to
distribute the daily precipitation amongst the timesteps in that day using a cascade process, or
allow the user to specify a normalised diurnal profile for each gridbox.
c Crown Copyright 2014 12
6 Acknowledgements
Many thanks to Pete Falloon, Martin Best and Andy Wiltshire for reading through various stages of
the draft and their helpful advice and suggestions. We are also very grateful to Spencer Liddicoat,
Camilla Mathison and James Manners for their assistance with the Unified Model set-up.
References
M. J. Best, M. Pryor, D. B. Clark, G. G. Rooney, Essery, C. B. Menard, J. M. Edwards, M. A.
Hendry, A. Porson, N. Gedney, L. M. Mercado, S. Sitch, E. Blyth, O. Boucher, P. M. Cox, C. S. B.
Grimmond, and R. J. Harding. The Joint uk Land Environment Simulator (jules), model description
Part 1: Energy and water fluxes. Geoscientific Model Development, 4(3):677699, September
2011. doi: 10.5194/gmd-4-677-2011. URL http://dx.doi.org/10.5194/gmd-4-677-2011.
D. B. Clark, L. M. Mercado, S. Sitch, C. D. Jones, N. Gedney, M. J. Best, M. Pryor, G. G. Rooney,
R. L. H. Essery, E. Blyth, O. Boucher, R. J. Harding, C. Huntingford, and P. M. Cox. The Joint
uk Land Environment Simulator (jules), model description Part 2: Carbon fluxes and vegetation
dynamics. Geoscientific Model Development, 4(3):701722, September 2011. doi: 10.5194/gmd-
4-701-2011. URL http://dx.doi.org/10.5194/gmd-4-701-2011.
Emma Compton and Martin Best. Impact of spatial and temporal resolution on modelled terrestrial
hydrological cycle components. Technical Report 44, WATCH report, July 2011.
Rutger Dankers, Nigel W. Arnell, Douglas B. Clark, Pete D. Falloon, Balazs M. Fekete, Simon N.
Gosling, Jens Heinke, Hyungjun Kim, Yoshimitsu Masaki, Yusuke Satoh, Tobias Stacke, Yoshihide
Wada, and Dominik Wisser. First look at changes in flood hazard in the Inter-Sectoral impact
model intercomparison project ensemble. Proceedings of the National Academy of Sciences,
111(9):2013020783261, December 2013. ISSN 1091-6490. doi: 10.1073/pnas.1302078110.
URL http://dx.doi.org/10.1073/pnas.1302078110.
c Crown Copyright 2014 13
J. C. S. Davie, P. D. Falloon, R. Kahana, R. Dankers, R. Betts, F. T. Portmann, D. B. Clark, A. Itoh,
Y. Masaki, K. Nishina, B. Fekete, Z. Tessler, X. Liu, Q. Tang, S. Hagemann, T. Stacke, R. Pavlick,
S. Schaphoff, S. N. Gosling, W. Franssen, and N. Arnell. Comparing projections of future changes
in runoff and water resources from hydrological and ecosystem models in ISI-MIP. Earth System
Dynamics Discussions, 4(1):279315, February 2013. ISSN 2190-4995. doi: 10.5194/esdd-4-
279-2013. URL http://dx.doi.org/10.5194/esdd-4-279-2013.
R. Essery, M. Best, and P. Cox. Moses 2.2 technical documentation. Technical Report 30, Hadley
Centre, 2001. URL http://www.metoffice.gov.uk/media/pdf/9/j/HCTN 30.pdf.
Chris Hewitt, Carlo Buontempo, and Paula Newton. Using climate predictions to better serve soci-
etys needs. Eos Trans. AGU, 94(11):105107, March 2013. doi: 10.1002/2013eo110002. URL
http://dx.doi.org/10.1002/2013eo110002.
C. Huntingford, B. B. B. Booth, S. Sitch, N. Gedney, J. A. Lowe, S. K. Liddicoat, L. M. Mercado,
M. J. Best, G. P. Weedon, R. A. Fisher, M. R. Lomas, P. Good, P. Zelazowski, A. C. Everitt, A. C.
Spessa, and C. D. Jones. IMOGEN: an intermediate complexity model to evaluate terrestrial
impacts of a changing climate. Geoscientific Model Development, 3(2):679687, November 2010.
doi: 10.5194/gmd-3-679-2010. URL http://dx.doi.org/10.5194/gmd-3-679-2010.
Graeme L. Stephens, Tristan LEcuyer, Richard Forbes, Andrew Gettlemen, Jean-Christophe Golaz,
Alejandro Bodas-Salcedo, Kentaroh Suzuki, Philip Gabriel, and John Haynes. Dreary state of
precipitation in global models. J. Geophys. Res., 115(D24):D24211+, December 2010. ISSN
0148-0227. doi: 10.1029/2010jd014532. URL http://dx.doi.org/10.1029/2010jd014532.
A. J. Stirling and R. A. Stratton. Entrainment processes in the diurnal cycle of deep convection
over land. Q.J.R. Meteorol. Soc., 138(666):11351149, July 2012. doi: 10.1002/qj.1868. URL
http://dx.doi.org/10.1002/qj.1868.
R. A. Stratton and A. J. Stirling. Improving the diurnal cycle of convection in GCMs.
Q.J.R. Meteorol. Soc., 138(666):11211134, July 2012. doi: 10.1002/qj.991. URL
http://dx.doi.org/10.1002/qj.991.
D. N. Walters, K. D. Williams, I. A. Boutle, A. C. Bushell, J. M. Edwards, P. R. Field, A. P. Lock, C. J.
Morcrette, R. A. Stratton, J. M. Wilkinson, M. R. Willett, N. Bellouin, A. Bodas-Salcedo, M. E.
Brooks, D. Copsey, P. D. Earnshaw, S. C. Hardiman, C. M. Harris, R. C. Levine, C. MacLachlan,
J. C. Manners, G. M. Martin, S. F. Milton, M. D. Palmer, M. J. Roberts, J. M. Rodrguez, W. J.
Tennant, and P. L. Vidale. The Met Office Unified Model Global Atmosphere 4.0 and jules Global
Land 4.0 configurations. Geoscientific Model Development Discussions, 6(2):28132881, May
2013. doi: 10.5194/gmdd-6-2813-2013. URL http://dx.doi.org/10.5194/gmdd-6-2813-2013.
Lila Warszawski, Katja Frieler, Veronika Huber, Franziska Piontek, Olivia Serdeczny, and
Jacob Schewe. The Inter-Sectoral impact model intercomparison project (ISIMIP):
c Crown Copyright 2014 14
Project framework. Proceedings of the National Academy of Sciences, pages
201312330+, December 2013. ISSN 1091-6490. doi: 10.1073/pnas.1312330110. URL
http://dx.doi.org/10.1073/pnas.1312330110.
G. P. Weedon, S. Gomes, P. Viterbo, W. J. Shuttleworth, E. Blyth, H. Osterle, J. C. Adam,
N. Bellouin, O. Boucher, and M. Best. Creation of the WATCH forcing data and its use to
assess global and regional reference crop evaporation over land during the twentieth cen-
tury. J. Hydrometeor, 12(5):823848, March 2011. doi: 10.1175/2011jhm1369.1. URL
http://dx.doi.org/10.1175/2011jhm1369.1.
A Additional plots: global climatology plots and diurnal cycles
of forcing variables at FLUXNET sites
c Crown Copyright 2014 15
18
0
120
60
060
120
180
60
3003060
soil
evap
otra
nspi
rati
on
0.0
0.6
1.2
1.8
2.4
3.0
3.6
4.2
mm
/day
18
0
120
60
060
120
180
60
3003060
soil
evap
otra
nspi
rati
onan
omal
y,3h
rly,
inte
rpol
ated
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
18
0
120
60
060
120
180
60
3003060
soil
evap
otra
nspi
rati
onan
omal
y,di
sagg
,conv
rain
=6h
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
18
0
120
60
060
120
180
60
3003060
soil
evap
otra
nspi
rati
onan
omal
y,da
ily,n
odi
sagg
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
Figu
re7:
Top
left:
Out
putv
aria
ble
from
JULE
Sru
nw
ithfu
ll20
min
ute
reso
lutio
nfo
rcin
g.To
prig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
g3
hour
lyfo
rcin
gw
ithin
terp
olat
ion.
Bot
tom
left:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gdi
sagg
rega
tion
with
defa
ulte
vent
dura
tions
.B
otto
mrig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gda
ilyfo
rcin
gw
ithno
inte
rpol
atio
n.
c Crown Copyright 2014 16
18
0
120
60
060
120
180
60
3003060
evap
from
cano
py/s
urfa
cest
ore
0.0
0.4
0.8
1.2
1.6
2.0
2.4
2.8
mm
/day
18
0
120
60
060
120
180
60
3003060
evap
from
cano
py/s
urfa
cest
ore
anom
aly,
3hrl
y,in
terp
olat
ed
1.
8
1.2
0.
60.
00.
61.
21.
8
mm
/day
18
0
120
60
060
120
180
60
3003060
evap
from
cano
py/s
urfa
cest
ore
anom
aly,
disa
gg,
conv
rain
=6h
1.
8
1.2
0.
60.
00.
61.
21.
8
mm
/day
18
0
120
60
060
120
180
60
3003060
evap
from
cano
py/s
urfa
cest
ore
anom
aly,
daily
,no
disa
gg
1.
8
1.2
0.
60.
00.
61.
21.
8
mm
/day
Figu
re8:
Top
left:
Out
putv
aria
ble
from
JULE
Sru
nw
ithfu
ll20
min
ute
reso
lutio
nfo
rcin
g.To
prig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
g3
hour
lyfo
rcin
gw
ithin
terp
olat
ion.
Bot
tom
left:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gdi
sagg
rega
tion
with
defa
ulte
vent
dura
tions
.B
otto
mrig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gda
ilyfo
rcin
gw
ithno
inte
rpol
atio
n.
c Crown Copyright 2014 17
18
0
120
60
060
120
180
60
3003060
surf
ace
runo
ff
0.0
0.8
1.6
2.4
3.2
4.0
4.8
mm
/day
18
0
120
60
060
120
180
60
3003060
surf
ace
runo
ffan
omal
y,3h
rly,
inte
rpol
ated
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
18
0
120
60
060
120
180
60
3003060
surf
ace
runo
ffan
omal
y,di
sagg
,conv
rain
=6h
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
18
0
120
60
060
120
180
60
3003060
surf
ace
runo
ffan
omal
y,da
ily,n
odi
sagg
0.
9
0.6
0.
30.
00.
30.
60.
9
mm
/day
Figu
re9:
Top
left:
Out
putv
aria
ble
from
JULE
Sru
nw
ithfu
ll20
min
ute
reso
lutio
nfo
rcin
g.To
prig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
g3
hour
lyfo
rcin
gw
ithin
terp
olat
ion.
Bot
tom
left:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gdi
sagg
rega
tion
with
defa
ulte
vent
dura
tions
.B
otto
mrig
ht:
Ano
mal
ies
with
resp
ectt
oth
efu
llre
sult,
usin
gda
ilyfo
rcin
gw
ithno
inte
rpol
atio
n.
c Crown Copyright 2014 18
18
0
120
60
060
120
180
60
3003060
sub
surf
ace
runo
ff
0.0
0.4
0.8
1.2
1.6
2.0
2.4
2.8
mm
/day
18
0
120
60
060
120
180
60
3003060
sub
surf
ace
runo
ffan
omal
y,3h
rly,
inte
rpol
ated
0.
48
0.32
0.
160.
000.
160.
320.
48
mm
/day
18
0
120
60
060
120
180
60
3003060
sub
surf
ace
runo
ffan
omal
y,di
sagg
,conv
rain
=6h
0.
48
0.32
0.
160.
000.
160.
320.
48
mm
/day
18
0
120
60
060
120
180
60
3003060
sub
surf
ace
runo
ffan
omal
y,da
ily,n
odi
sagg
0.
48
0.32
0.
160.
000.
160.
320.
48
mm
/day
Figu
re10
:Top
left:
Out
putv
aria
ble
from
JULE
Sru
nw
ithfu
ll20
min
ute
reso
lutio
nfo
rcin
g.To
prig
ht:A
nom
alie
sw
ithre
spec
tto
the
full
resu
lt,us
ing
3ho
urly
forc
ing
with
inte
rpol
atio
n.B
otto
mle
ft:A
nom
alie
sw
ithre
spec
tto
the
full
resu
lt,us
ing
disa
ggre
gatio
nw
ithde
faul
teve
ntdu
ratio
ns.
Bot
tom
right
:A
nom
alie
sw
ithre
spec
tto
the
full
resu
lt,us
ing
daily
forc
ing
with
noin
terp
olat
ion.
c Crown Copyright 2014 19
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=3h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=6h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=12h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=24h
0.9 0.6 0.3 0.0 0.3 0.6 0.9mm/day
soil evapotranspiration
Figure 11: Anomalies with respect to the full result, using disaggregation with convective rain eventdurations of 3 hours (top left), 6 hours (top right), 12 hours (bottom left) and 24 hours (bottom right).
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=3h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=6h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=12h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=24h
1.8 1.2 0.6 0.0 0.6 1.2 1.8mm/day
evap from canopy/surface store
Figure 12: Anomalies with respect to the full result, using disaggregation with convective rain eventdurations of 3 hours (top left), 6 hours (top right), 12 hours (bottom left) and 24 hours (bottom right).
c Crown Copyright 2014 20
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=3h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=6h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=12h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=24h
0.9 0.6 0.3 0.0 0.3 0.6 0.9mm/day
surface runoff
Figure 13: Anomalies with respect to the full result, using disaggregation with convective rain eventdurations of 3 hours (top left), 6 hours (top right), 12 hours (bottom left) and 24 hours (bottom right).
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=3h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=6h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=12h
180 120 60 0 60 120 180
60
30
0
30
60
disagg, conv rain
=24h
0.48 0.32 0.16 0.00 0.16 0.32 0.48mm/day
sub surface runoff
Figure 14: Anomalies with respect to the full result, using disaggregation with convective rain eventdurations of 3 hours (top left), 6 hours (top right), 12 hours (bottom left) and 24 hours (bottom right).
c Crown Copyright 2014 21
Figure 15: Diurnal cycle of driving data for the Hyytiala FLUXNET site, Finland, averaged over thedays in 1984.
c Crown Copyright 2014 22
Figure 16: Diurnal cycle of driving data for the Vielsalm FLUXNET site, Belgium, averaged over thedays in 1984.
c Crown Copyright 2014 23
Figure 17: Diurnal cycle of driving data for the Collelongo FLUXNET site, Italy, averaged over thedays in 1984.
c Crown Copyright 2014 24
Figure 18: Diurnal cycle of driving data for the Harvard Forest FLUXNET site, Massachusetts,averaged over the days in 1984.
c Crown Copyright 2014 25
Figure 19: Diurnal cycle of driving data for the Bondville FLUXNET site, Illinois, averaged over thedays in 1984.
c Crown Copyright 2014 26
Met Office Tel: 0870 900 0100
FitzRoy Road, Exeter Fax: 0870 900 5050
Devon, EX1 3PB [email protected]
UK www.metoffice.gov.uk
Produced with the Met Office LaTeX template v2.0.0